metehan.ai Research & SOPs

Appearance

Fresh

Recency Bias in AI Search ​

Freshness is a confirmed ranking factor across all major AI search platforms. Newer content gets preferential treatment in citation selection, and the effect is measurable. This page summarizes the evidence from Metehan's research and cross-references findings from other articles in this collection.

The 125K Impressions Experiment ​

Metehan tested recency bias by publishing content on the same topic at different time intervals and measuring AI search citation rates. The experiment generated over 125,000 AI search impressions across ChatGPT, Perplexity, and Google AI Mode.

Key findings:

  • Content published within 7 days consistently outperformed identical content published 30+ days ago
  • The freshness advantage was strongest in ChatGPT (estimated 15-25% citation boost for content < 7 days old)
  • Perplexity showed moderate recency preference (estimated 5-15% boost)
  • Google AI Mode applied freshness variably by query type (high for news, low for evergreen)

Code-Level Confirmation ​

The freshness signal is not inferred from behavior alone. It is confirmed in the actual system configurations:

ChatGPT ​

  • freshness_weight parameter observed in search configuration (see ChatGPT Search Configuration)
  • Exponential freshness decay function
  • freshness_override boolean for breaking news queries

Google AI Mode ​

  • Freshness is Signal 6 of the 7-signal ranking stack (see Google AI Mode Architecture)
  • Freshness weight varies by query type (breaking news = very high, evergreen = low)
  • Uses datePublished, dateModified, HTTP headers, and content-level temporal signals

Perplexity ​

  • Recency preference is moderate compared to ChatGPT
  • Evergreen content can still compete with newer content if sufficiently comprehensive
  • Topic multipliers interact with freshness (news topics have higher freshness weight)

Google Discover ​

  • 6-tier freshness bucketing system (see Google Discover Architecture)
  • Real-time content (0-4 hours) gets highest priority
  • Content over 30 days moves to archived status with minimal feed inclusion

Practical Implications ​

Freshness is Not a Shortcut

Updating the dateModified tag without changing content does not fool these systems. All platforms use content-level freshness signals (new data points, updated statistics, current-year references) in addition to metadata dates. Fake freshness is detectable and can hurt trust signals.

What Works ​

  1. Regular content updates with genuine new information. Add new data, reference recent events, update statistics. The freshness signal rewards real updates.

  2. Publish fast on trending topics. For news and emerging topics, the freshness decay is steep. Being first matters.

  3. Maintain evergreen content separately. Not all content benefits from freshness chasing. Comprehensive evergreen guides compete on depth, not recency. Update them quarterly with substantive additions.

  4. Use schema markup accurately. datePublished and dateModified are read by all platforms. Accuracy builds trust; manipulation erodes it.

What Does Not Work ​

  • Changing dates without changing content
  • Republishing the same article with a new URL
  • Minor edits (fixing typos, changing a word) to trigger a modified date
  • Automated content refreshes that add no substantive value

Cross-Reference

For platform-specific freshness optimization, see: